What about updating the exclusion file instead?

Build is green
See https://jenkins.softwareheritage.org/job/DJNL/job/tox/345/ for more details.

In D2757#65740, @vlorentz wrote:

What about updating the exclusion file instead?

That's not an either/or affair IMO. By definition as long as the replayer is running, the exclusion file is always out of date (and updating it is quite costly).

I also don't know where the code used to generate the exclusion file is...

In D2757#65748, @olasd wrote:

That's not an either/or affair IMO. By definition as long as the replayer is running, the exclusion file is always out of date (and updating it is quite costly).

But if it doesn't matter if it doesn't see the same objects too many times

I also don't know where the code used to generate the exclusion file is...

https://forge.softwareheritage.org/source/snippets/browse/master/vlorentz/objstorage-replay-exclusion-file/

but you have to ask @seirl to generate unsorted_inventory.txt.gz for you

In D2757#65755, @vlorentz wrote:

In D2757#65748, @olasd wrote:

That's not an either/or affair IMO. By definition as long as the replayer is running, the exclusion file is always out of date (and updating it is quite costly).

But if it doesn't matter if it doesn't see the same objects too many times

I really can't tell if it matters or not until I have a knob I can turn on/off to compare. In any cases, it certainly doesn't hurt to have the option.

I also don't know where the code used to generate the exclusion file is...

https://forge.softwareheritage.org/source/snippets/browse/master/vlorentz/objstorage-replay-exclusion-file/

but you have to ask @seirl to generate unsorted_inventory.txt.gz for you

So this diff is somewhat needed anyway. We're currently excluding around 5-10% of writes, which is substantial (although, overall, we're currently slower than the fastest we've been in the past).